Similarity Learning in GIS: An Overview of Definitions, Prerequisites and Challenges

Abstract

In this chapter we review similarity learning in spatial databases. Traditional exact-match queries do not conform to the exploratory nature of GIS datasets. Non-adaptable query methods fail to capture the highly diverse needs, expertise and understanding of users querying for spatial datasets. Similarity-learning algorithms provide support for user preference and should therefore be a vital part in the communication process of geospatial information. More specifically, we address machine learning as applied in the optimization of query similarity. We review appropriate definitions of similarity and we position similarity learning within data mining and machine learning tasks. Furthermore, we outline prerequisites for similarity learning techniques based on the unique characteristics of the GIS domain. A description of specific methodologies follows based on the highly diverse attributes of GIS datasets (for example, text, images, video), and application examples are presented. We summarize previously set requirements and present future trends expected to emerge in the coming years.